Relative superintelligence and competitive dynamics: between collapse, progress, and breakthroughs

The key difficulty in modeling the impact of AI on competitive dynamics is that it induces an order of magnitude change in the potential information density of outputs — a term we'll use to represent the full depth of domain knowledge, contextual information, and algorithmic power embedded in any given output — large and sudden enough that structural limits in the information processing capabilities of markets and producers becomes significant.

To make these dynamics concrete we run a series of simulated market scenarios using demand and supply curves with the standard relationships to quality — we are interested here on information density — and cost. We will also assume markets large or fragmented enough that we can study in isolation three types of producers defined by the upper bound of the information density/complexity of their outputs (although this upper bound need not be the output complexity at equilibrium):

• Generic producers with an upper bound on the information density of their outputs that's relatively low compared to the maximum information density the market can evaluate.
• Expert producers who can generate outputs essentially as complex as the market is capable to understand.
• Augmented producers with, in this context, no upper bound to the possible complexity of their outputs.

This taxonomy allows us to represent differences in institutional, cultural, and organizational capabilities. It also gives us a workable, precise concept of superintelligence: not "above human intelligence" but simply above what a given market can evaluate and reward.

We show below equilibriums for all three types of producers under baseline cost and demand functions. There's no difference between expert and augmented producers. While the market is willing to pay more for information density above what a generic producer can offer, it can't recognize the even higher information density augmented producers are capable of. Augmented producers generate at equilibrium outputs with information complexity far below their potential.

Before looking at the impact of AI it's worth looking closer at a subset of markets with an even lower upper bound on information complexity demand. We've chosen to label markets where the upper bound of rewarded information density is comparable to the upper bound of generic producers as platform markets. This isn't meant to imply that there's no skill involved in selling in those markets but it points to the following empirical characteristics:

• Demand is driven more by the platform's logic than by individual choice (or at least the platform logic constrains and handicaps the set of offers to the point that individual choice is in a deep sense irrelevant).
• The logic of most if not all platforms is neutral in terms of information complexity, when not hostile to it.

Under these conditions equilibriums for all producer types coincide. Again, we aren't stating that there's no return to skill or capital in platform markets, just that there's no return to the information complexity in their outputs.

Platform markets — most visibly social networks — were the first ones to be fully exposed to changes in AI capabilities during the last few years and they are among the most socially and politically relevant so it makes sense to model first the impact on them.

For our purposes we model the impact of AI on production costs through these two characteristics:

• Compared to baseline production technology, costs are lowered at every information complexity.
• Unlike baseline production technology, costs rise more slowly as the information complexity of output grows, and this slope doesn't change at higher levels of information complexity.

(We are concerned here with properly deployed AI used in appropriate contexts. AI technically or conceptually misused induces large economic and sociopolitical costs outside the scope of this analysis.)

Changes to the cost function with respect to information complexity are the most significant. First, the introduction of AI makes it cheaper to add more information complexity to it through more intensive deployment of domain expertise and bespoke code. Second, these enhancements don't grow significantly more expensive at higher levels of information complexity as long as it's below the producer's own intrinsic organizational, cultural, and expertise limits.

Introducing AI to a platform market has a recognizable impact:

The qualitative explanation is straightforward. As platform markets don't reward medium or high-complexity outputs, all equilibriums remain low-complexity. But AI has made their cost lower leading to a general price collapse. This dynamic is driven by the interaction between AI-driven cost reduction and the shape of the demand function in platform markets. In the context of e.g. social networks, the fact that AI capabilities exceed the production of slop is irrelevant.

Baseline markets more sensitive to information complexity show a different response to the introduction of AI. Prices still fall dramatically for all outputs yet because the market rewards information complexity overall levels remain. In fact the new equilibrium can be not only at a lower cost but at a higher complexity, essentially the maximum the market is able to understand and reward.

There's yet a third type of market. We label it unbounded as the demand, as a function of information quality, has no immediately relevant upper bound. In these markets higher information complexity always increases the market value of the output, making upper bounds to producer's capabilities significant. High-end industrial technology and biotechnology are two examples of this sort of market. Unlike memes or listicles, the information complexity of generic, expert, and augmented producers in those markets is recognized and rewarded.

It's in those markets where the deeper capabilities of AI have the most impact. The equilibriums of generic and expert producers shift to lower costs, and of augmented producers to unprecedentedly complex outputs commanding a price differential.

The diversity of views on the impact of AI is driven by the underappreciated interaction between producer and market types. An expert producer in a platform market — a newspaper publishing on a social network — and an augmented producer in an unbounded-demand market — a biotechnology company tackling an unsolved medical problem — are going to see competitive dynamics change in very different ways.

The optimal strategic response depends on an accurate diagnosis of the relevant market and of the organization's own producer type. The latter can, to some degree, be changed, but not without the sort of radical transformation that's almost indistinguishable from the setup of a new organization.